Modelling the influence of individual level status discrimination on group level differences in chronic stress. Agent based model approach

(1)

U

NIVERSITEIT VAN

A

MSTERDAM

& V

RIJE

U

NIVERSITEIT

A

MSTERDAM

(JOINT DEGREE)

M

ASTER

T

HESIS

Modelling the influence of individual level

status discrimination on group level

differences in chronic stress

An Agent-Based model approach

Author:

Bsc. Dirk ZOMERDIJK

Supervisors: Dr. Nadège MERABET

Msc. Loes CRIELAARD

A thesis submitted in fulfillment of the requirements for the degree of Master of Science

in the

Computational Science

(2)

(3)

iii

Declaration of Authorship

I, Bsc. Dirk ZOMERDIJK, declare that this thesis titled, “Modelling the influence of

individ-ual level status discrimination on group level differences in chronic stress

An Agent-Based model approach” and the work presented in it are my own. I confirm that: • This work was done wholly or mainly while in candidature for a research degree at

this University.

• Where any part of this thesis has previously been submitted for a degree or any other qualification at this University or any other institution, this has been clearly stated. • Where I have consulted the published work of others, this is always clearly

at-tributed.

• Where I have quoted from the work of others, the source is always given. With the exception of such quotations, this thesis is entirely my own work.

• I have acknowledged all main sources of help.

• Where the thesis is based on work done by myself jointly with others, I have made clear exactly what was done by others and what I have contributed myself.

Signed:

(4)

(5)

v

UNIVERSITEIT VAN AMSTERDAM & VRIJE UNIVERSITEIT AMSTERDAM(JOINT DEGREE)

Abstract

Health Behaviors & Chronic Diseases Public Health

Master of Science

Modelling the influence of individual level status discrimination on group level differences in chronic stress

An Agent-Based model approach

by Bsc. Dirk ZOMERDIJK

Multiple studies have related lower socioeconomic status (SES) with experience of greater chronic stress. One specific aspect of chronic stress that may be relevant to explain socioe-conomic inequalities in health is psychosocial stress resulting from status discrimination. Traditional statistical models describing the relation between SES and stress that incor-porate a set of statistical assumptions do not take into account the complexity of human behaviour at the individual level. Traditional statistical models often assume causality, ignoring feedback loops that may exist between segregated pathways. To isolate and re-search the mechanisms influencing SES discrimination, we, therefore, propose to use an agent-based model (ABM). The identified mechanisms in itself, although simplified, were able to produce a gradient of stress over SES. Psychosocial resources (PSRs) may play an important role in explaining why some individuals are more affected by stress than others. Furthermore, we showed how the justification of lower status positions, by the higher sta-tus individuals, may contribute to the existence of stasta-tus inequality and in turn, raises the amount of stress in the system. Lastly, the sensitivity analysis revealed that interactions between the mechanisms explained more variance on the model outcome than the mech-anisms themselves. To conclude, we were able to identify the concepts that are relevant to status discrimination and performed the first step in building an ABM that can reproduce the hypothesized outcomes found in the literature. Through the explicit modelling of the dynamic processes involved in status discrimination, this modelling approach may inform how inequalities in experienced stress emerge, and how they may be reduced. The current model may serve a basis for more complex models and can be characterized numerically when the data is available. We describe which data would greatly improve the numerical implementation of the model, and suggest a research method on how to obtain this data.

(6)

(7)

vii

Acknowledgements

I cannot express enough thanks to my supervisors, Loes Crielaard and Dr. Nadege Mer-abet, for their continued guidance, and encouragement throughout the whole project. I would like to thank Dr. Rick Quax for being my examinator on the project and for his insights when things got (too) complex.

I want to show my gratitude for the board of the HELIUS study, and in specific Prof. Karien Stronks for providing me with the chance to work with them at the AMC depart-ment of Public health, for repeatably showing interest in the project and for reviewing the final draft. Also, I would like to thank the Institute of Advanced Studies for inviting me to take part in their excellent student program.

My completion of this project could not have been accomplished without the emotional support of my roommates, friends and family. A special thanks for Mannus, who did share not only his knowledge and writing skills with me but also his 24-core beast of a computer which saved me weeks of simulation time.

Finally, for my loving and supportive girlfriend, Lot: my deepest gratitude. Your en-couragement when times got rough is much appreciated and duly noted. It was of great comfort and relief to know you were completely supportive during this long and time consuming project. Now it is our time to shine. My heartfelt thanks.

(8)

(9)

ix

2.9.1 Emergence . . . 24 2.9.2 Adaption. . . 24 2.9.3 Objectives . . . 24 2.9.4 Learning . . . 24 2.9.5 Prediction . . . 24 2.9.6 Sensing. . . 24 2.9.7 Interaction . . . 25 2.9.8 Stochasticity . . . 25 2.9.9 Collectives . . . 25 2.9.10 Observation . . . 25 2.10 Outcome measures . . . 25 2.11 Implementation . . . 25 2.12 Prior analysis . . . 26 2.13 Coefficient of variation . . . 26 2.14 Validity . . . 27 2.14.1 Face validity . . . 27

2.14.2 Internal and external validity . . . 27

2.14.3 Statistical integrity . . . 28

2.15 Sensitivity analysis . . . 28

2.15.1 Local sensitivity (OFAT) . . . 28

2.15.2 Global sensitivity (Sobol) . . . 28

3 Results 31 3.1 Validity . . . 31

3.1.1 Initial face validity . . . 31

3.1.2 Statistic integrity . . . 31

3.1.3 Internal validity . . . 31

3.2 Sensitivity analysis . . . 33

3.2.1 Local sensitivity (OFAT) . . . 33

3.2.2 Global sensitivity (SOBOL) . . . 35

4 Discussion and future work 43 4.1 Evaluation of the findings . . . 43

4.1.1 Distribution of PSRs . . . 43

4.1.2 Sensitivity analyses. . . 43

OFAT . . . 43

Sobol . . . 44

(11)

xi

4.3 Limitations . . . 45

4.4 Future work . . . 47

4.5 Concluding summary . . . 48

(12)

(13)

xiii

List of Figures

2.1 Function for translating status difference to perceived stressor . . . 19

2.2 Function for translating accumulated stress to vulnerability . . . 19

2.3 Function for determining the prestige growth rate coefficient . . . 20

2.4 The ABM model. The processes indicated with P#, represent processes that are equal for all agents. Processes indicated with I# are processes that differ between agents depending on resources or previous interactions. The loops indicated with F# represent the feedback loops driven by the sequential pro-cesses.. . . 23

2.5 Coeeficient of variation for increasing numbers of repetitions . . . 27

3.1 Distribution of average stress for 1000 model simulations with the default parameter set. . . 31

3.2 The relation between SES and experienced stress. The x-axis represents SES, the y-axis represents the amount of stress at the end of simulation. The color of the circles indicate the PSRs. For each status group (1-14), the amount of PSRs is normalized such that 0 are the agents with the lowest PSRs in that status group and 1 are the agents with the highest PSRs. . . 32

3.3 Average stress of population grouped in three status categories. Left is low status (SES <= 5), middle is medium (5 < SES <= 10) and right is high (SES > 10). Within each status category the population is divided in three PSRs categories, low (red, PSRs <= 0.33), medium (blue, 0.33 < PSRs <= 0.66), and high (green, PSRs > 0.66). . . 33

3.4 Average PSRs for each status group. . . 33

3.5 Distribution of prestige over status groups. . . 34

3.6 Timeseries of prestige gain for status groups in the low status (A, red), medium status (B, blue) and high status categories (C, green). . . 34

3.7 Results OFAT: coping noise. . . 34

3.8 Results OFAT: number of interactions. . . 35

3.9 Results OFAT: prestige coefficient. . . 35

3.10 Results OFAT: prestige growth rate.. . . 36

3.11 Results OFAT: PSRs coefficient. . . 36

3.12 Results OFAT: recovery rate. . . 37

3.13 Results OFAT: status estimation noise. . . 37

3.14 Results OFAT: baseline similarity. . . 38

3.15 Results OFAT: stressor coefficient.. . . 38

3.16 Results OFAT: vulnerability coefficient. . . 38

3.17 First order and Total order indices for the outcome measure average stress in system. The circles represent the average amount of explained variance, the errorbars represent the 95% confidence bounds on this estimation . . . . 39

3.18 First order and Total order indices for the steepness of the SES - stress gra-dient. The circles represent the average amount of explained variance, the errorbars represent the 95% confidence bounds on this estimation . . . 39

(14)

xiv

3.19 First order and Total order indices for the steepness of the SES - stress gra-dient. The circles represent the average amount of explained variance, the errorbars represent the 95% confidence bounds on this estimation . . . 40

3.20 First order and Total order indices for the steepness of the SES - prestige gradient. The circles represent the average amount of explained variance, the errorbars represent the 95% confidence bounds on this estimation . . . . 40

3.21 Second order indices for the average stress in the system and the slope of the SES - stress gradient . . . 41

(15)

xv

List of Tables

2.1 Variables of HELIUS dataset. . . 13

2.2 Distribution of ethnicities in Amsterdam. . . 14

2.3 Occupation composite scores . . . 15

2.4 SES composite scores for: education, occupation, and income difficulty. . . . 15

(16)

(17)

xvii

List of Abbreviations

SES Socioeconomic Status

USA United Sstates of America

PSR Psychosocial Rersource

ABM Agent Based Model

HELIUS HEalthy Life in a Urban Setting

DES Daily Emotional Support

SSQT Docial Eupport Questionnaire for Transactions

SSQS Docial Eupport Questionnaire for Supportive transactions

CBS Centraal Bureau voor de Statistiek

LHS Latin Hypercube Sampling

(18)

(19)

xix

List of Symbols

A Agent N Population size T Time index ∆t Time-step size As Agent SES Ar Agent PSRs E Estimates status

Si Agent accumulated stress

Ach Agent chronic stressed state

Av Agent vulnerability

Gs Status group

Gn Members in status group

Gpr Prestige status group

s Stressor τ Chronic threshold σ Baseline similarity εs Status estimation ρ PSRs coefficient εc Coping noise Rβ Recovery rate ξ Stressor coefficient υ Vulnerability coefficient ψ Prestige coefficient

(20)

(21)

1

Chapter 1

Background and research goals

1.1 The socioeconomic status – health gradient

Socioeconomic status (SES) describes an individual’s relative position in society measured by social and economic markers such as wealth, educational attainment and occupational level. Numerous studies have shown that inequality in health over SES groups is signifi-cant (McGowan and Shahab,2019). Epidemiological studies consistently recognize graded relations between SES and major physical health outcomes, including hypertension, car-diovascular disease and all-cause mortality (Major, Mendes, and Dovidio,2013; Wilkinson and Pickett,2009; Spruill,2010). These socioeconomic inequalities in health have been re-searched for decades, and although interventions are made to reduce these inequalities, they remained persistent (Mackenbach,2012). All countries, whether with relatively weak (e.g. the United States of America (USA)) or stronger (e.g. the Netherlands) welfare sys-tems, have systematic inequalities in mortality and morbidity between residents with a higher and a lower SES (Galama and Van Kippersluis, 2019; Mackenbach,2012). For ex-ample, a study in the USA showed that a 20-year-old low-income male, on average, reports being similar in health as a 60-year-old high-income male (Case and Deaton, 2005). This recurring and robust positive association between SES and health is commonly referred to as the SES-health gradient (Galama and Van Kippersluis,2019).

1.2 Mechanisms related to SES - health gradient

Multiple mechanisms relating SES to health have been proposed (Adler et al.,1994; Mack-enbach, 2012). Some intuitive mechanisms for explaining the gradient are factors related to behaviour, to quality of healthcare and to the physical and social environment.

1.2.1 Behaviour

In the context of factors related to behaviour, research, for instance, found the existence of a socioeconomic gradient in diet (McLaren,2007), where individuals of higher SES groups tend to have a healthier diet, indicated by greater consumption of vegetables and fruit, and lower consumption of fats (Power, 2005). One reason provided for this difference among SES groups is that foods that make up a healthier diet are more expensive than less nutritious food items (McLaren,2007). Therefore, the economic capacity of lower SES groups may not be sufficient to provide proper access to nutritious foods. In this respect, low income communities may be vulnerable to obesity because (affordable) lower-cost foods tend to be high in energy, but low in nutrients, resulting in more obesity on average in the lower SES groups. Notably, income is not the only SES indicator that is an inverse correlate of obesity (McLaren,2007). Lower education and lower occupational status have also been associated with obesity (McLaren,2007; Drewnowski et al.,2014), indicating that considering other mechanisms is important.

(22)

2 Chapter 1. Background and research goals 1.2.2 Quality of healthcare

Research about healthcare suggests that physicians can hold biases that affect their re-sponses towards members of lower status groups within the medical context (Major, Mendes, and Dovidio,2013). Among these group members are black and obese individuals, as mul-tiple studies find that these characteristics are correlated with lower SES (Major, Mendes, and Dovidio, 2013; Hebl and Xu, 2001). For example, physicians explicitly communi-cate with a more negative approach towards obese than towards average-weight patients, and they report to spend less time treating obese persons (Hebl and Xu,2001). For other groups, such as Blacks, strong social norms exist against conveying bias, and healthcare providers are therefore less likely to express prejudice and stereotypes explicitly (Major, Mendes, and Dovidio, 2013). However, high levels of implicit bias remain towards these groups. These biases can systematically influence medical decision-making of healthcare providers (Major, Mendes, and Dovidio, 2013). For example, research found that white physicians who score higher on implicit racial bias were less likely to recommend appro-priate medical treatment to black than to white patients (Green et al.,2007).

1.2.3 Physical environment

In a study on the potential role of environmental risk exposure, researchers found inverse relations between income and other indices of SES with exposure to biological risk factors, including hazardous wastes and other toxins, ambient and indoor air pollutants, low wa-ter quality and ambient noise (Evans and Kantrowitz,2002). They also provide evidence that such exposures are harmful to health and well-being overall (Evans and Kantrowitz, 2002).

1.2.4 Social environment

Social factors such as residential crowding, and neighbourhood quality – where measure-ments include the amount of traffic, noise, and safe spaces -also have an inverse relation with indices of SES, and are known to influence health through various psychological pathways (Evans and Kantrowitz,2002). For example, residential crowding is related to more noise and more unwanted interactions, which are known to cause psychological distress and affect health (Evans and Kantrowitz, 2002). Neighbourhood quality is also related to increased exposure to noise pollution, and to crime and violence, which in turn affect various types of health problems (Evans and Kantrowitz, 2002). Moreover, expo-sure to these social factors has a negative effect on children’s socioemotional development (Evans and Kantrowitz,2002). Overall, low SES social environments have been related to increased psychological distress and impaired child socioemotional development, among others.

1.3 Mechanisms related to stress

In this regard, mechanisms related to greater exposure to and experience of greater chronic stress have also been proposed as explanations for the SES – health gradient (Adler et al., 1994; Major, Mendes, and Dovidio, 2013; Wilkinson, 1997; Wilkinson and Pickett, 2009). We define chronic stress, the variable of interest in this study, as a stressed state resulting from continuous accumulated stress over a time period longer than a week (Contrada and Baum,2010). We define stress according to the definition of Lazarus and Folkman (1984), who proposed that stress arises when individuals perceive that the demands of external situations are beyond their coping capacity. Furthermore, psychological stress research

(23)

1.4. Status discrimination and psychosocial stress 3 distinguished two aspects of stress: stressors and stress response (Lazarus and Folkman, 1984). Stressors, including the previously mentioned situations such as exposure to noise, interpersonal problems, and trauma, may induce stress. Stress may cause non-specific physical and mental changes which are called the stress response, which in turn may lead to stressed states such as frustration, depression, and anxiety, (Suzuki and Ito,2013).

Multiple studies have related lower SES with experience of greater chronic stress (Chen et al., 2006; Steptoe and Feldman, 2001; Spruill, 2010). Research has associated chronic stress with increased risk of disease, including cardiovascular disease, hypertension, and infectious disease (Cohen, Janicki-Deverts, and Miller,2007)

Over the past decades, research on the mechanisms relating SES to health were not able to fully explain the SES - health gradient (Adler et al.,1994; Wilkinson and Pickett,2009; Kogler et al.,2015). This may suggest that mechanisms exist that induce chronic stress in addition to the previously mentioned mechanisms.

1.4 Status discrimination and psychosocial stress

One specific aspect of chronic stress that may be relevant to explain socioeconomic in-equalities in health related to chronic stress is psychosocial stress. Psychosocial stress is stress caused by experiences of social threat, including discrimination, social evalua-tion and social exclusion (Kogler et al.,2015). Richard Wilkinson and Kate Prickett (2009) proposed a theory in which they explain why status inequality may induce psychosocial stress, and why reducing income differences in society may reduce status inequality and in turn eliminate many social problems and ills including psychosocial stress (Wilkinson and Pickett,2009). The reasoning behind this theory was that greater income equality re-duces individuals’ stress levels and prevents the associated health problems (Wilkinson and Pickett,2009).

In this regard, research has shown that mortality in developed countries is more re-lated to relative than absolute living standards in three ways (Wilkinson and Pickett,2009). Firstly, mortality is related more closely to relative income differences within countries than differences in absolute income between them (Wilkinson and Pickett, 2009). Sec-ondly, national mortality rates tend to be the lowest in countries that have smaller income differences (Wilkinson and Pickett, 2009). Thirdly, most of the long-term rise in life ex-pectancy seems unrelated to absolute long-term economic growth rates (Wilkinson and Pickett,2009). Although absolute material influences contribute to inequalities in health, the importance of relative standards that these findings support implies that psychosocial pathways may be particularly influential (Wilkinson and Pickett,2009).

This proposition is justified by linking two arguments. The first argument is that social problems observed at the level of society as a whole can be attributed to increased status competition among individual members of society (Wilkinson and Pickett,2009). Greater status inequality has also been correlated with increased numbers of homicide, violent crime, drug abuse, and obesity, among others (Wilkinson and Pickett,2009). Furthermore, the feeling of being in constant competition with others is related to the development of psychological symptoms of stress, which then can turn into physical illness over time. The second argument is the increased potential of "being judged" on the basis of status, which may be stressful (Wilkinson and Pickett,2009).

Thus, they argue, as inequality in a society increases, competition between individual members of the society increases. In addition, the possibility of being judged as a result of status is higher as granularity in status increases. In turn, these mechanisms trigger stress responses in individuals. Here, being judged on the basis of SES is termed socioeconomic status discrimination, or status discrimination (Van Dyke et al.,2016; Peterman,2018).

(24)

4 Chapter 1. Background and research goals In the following sections, we present the literature that describes how individuals eval-uate and communicate social status, the process of experiencing psychosocial stress as a result of status discrimination, and how individuals cope with this experience of stress.

1.5 Social status evaluation

The word “status” refers to a position in a social system occupied by actors that abide by a set of roles that define the involved members’ expected behaviours with members of related statuses (Schooler,2013). Statuses may be ranked in terms of the related concepts of 1) prestige, 2) access to, and chances to acquire relatively limited social resources, and 3) power, i.e. the capability to convince others to accomplish one’s objectives (Schooler, 2013). When statuses are treated hierarchically, they are often referred to as “social sta-tus” (Schooler,2013). In research, the classification of social status is often performed by measuring traditional SES indices– income, occupational level and educational attainment (Koski, Xie, and Olson,2015). However, previous studies have demonstrated that educa-tion, income and occupational status cannot be used interchangeably as indicators of an assumptive latent social dimension (Geyer et al.,2006; Festin et al.,2017).

1.5.1 SES measurements

Different measures of SES stand for access to different sorts of resources and prestige where each indicator embodies different aspects. For example, education grants access to knowledge. Income-based indicators involve access to material and immaterial resources for health, but also stand for a level of prestige. Occupational status promotes the per-ception of professional identity and community and is an influential marker for working conditions, both physical and psychosocial (Geyer et al., 2006; Shavers, 2007). Certainly, studies have shown that the effect of above mentioned SES measures varies depending on the health outcome under study. Therefore, despite their correlation, education, income and occupational status cannot be used interchangeably as indicators of the theoretical la-tent social variable SES, and thus measurements of SES should be based on a combination of these three indices (Major, Mendes, and Dovidio,2013).

1.5.2 Status perception

Research on person perception suggests that SES is related to a set of nonverbal cues, where higher SES individuals showed more disengagement and less engagement during interactions than did lower SES individuals (Kraus and Keltner,2009). Furthermore, naive observers of these interactions were able to identify the participants’ family income, mater-nal education, and subjective SES with above-chance accuracy, based on the participants’ non-verbal behaviour (Kraus and Keltner,2009). However, there are other, more obvious candidate status cues of SES such as appearance, style, and other behavioral tendencies (Kraus and Keltner,2009; Power,1999; Kraus, Park, and Tan,2017; McLaren,2007). 1.5.3 Communication of social status

Research suggests that “economic inequality is experienced daily as the communication of social class signals – behaviours that provide information about a person’s income, ed-ucational attainment, or occupational status – perceived and expressed in everyday in-teractions” (Kraus, Park, and Tan, 2017). Findings from psychology and economy were used to determine three basic predictions about social class signalling and the experience of inequality (Kraus, Park, and Tan,2017). They theorize that status signals 1) happen fast,

(25)

1.6. Experiencing status discrimination 5 frequently and accurately in the social perception process, 2) cause a glass ceiling effect between people of low and high status, and 3) generate psychological processes and be-haviours that legitimize and preserve the present socioeconomic system (Kraus, Park, and Tan,2017).

Moreover, research on social class signalling demonstrates that observers are able to make fast and accurate judgements about the social status of others using only small amounts of information (Kraus, Park, and Tan,2017). Theories on social status state that it shapes the lives of individuals by limiting or increasing access along dimensions of eco-nomic resources or by influencing behaviour via cultural and psychological mechanisms.

In this regard, social status influences multiple cultural aspects of an individual’s life, i.e. the food they eat, the music they listen to, how they spend their leisure time, the linguistic style they use, and the type of clothes they wear. This suggests that during social interactions, individuals’ behaviours and cultural practices are infused with their social status, and as a result, individuals accurately communicate their social status to each other (Kraus, Park, and Tan,2017).

Studies on social comparison – “the process of comparing oneself with others on so-cial characteristics and outcomes (Kraus, Park, and Tan,2017)” – illustrate that individuals compare themselves to others frequently on economic dimensions. Social comparison is regularly mentioned as “an unavoidable aspect of perception during social interactions (Kraus, Park, and Tan,2017)”, and is a mechanism used by individuals to obtain informa-tion about the self, regulate their emoinforma-tions and goals, manage uncertainty, and judge the normality of personal life events (Kraus, Park, and Tan,2017).

1.5.4 Prestige

Prestige is defined as "the respect and value that somebody/something has because of their social position, or what they have done" in the Oxford Advanced Learner’s Dic-tionary (2020). As the definition implies, prestige can be based on social position (e.g. SES), but can also be achieved through one’s actions. Prestige will influence interactions through the aforementioned status cues (Jury et al., 2019). These interactions can have consequences for opportunities and access to services, among others (Campbell, Marsden, and Hurlbert, 1986). Hence, discrimination in social interaction might, in turn, have a negative impact on the status of an individual. Prevention of access to opportunities and services can decrease SES by affecting economic and social resources. Moreover, the occur-rence of a discriminatory interaction against an individual might influence the perception of the discriminator or of other spectators into an amplifying global mechanism which will influence interactions with individuals considered to be “similar” (brewer2017intergroup, Falk and Zehnder, 2007). Finally, if discrimination impacts processes measured statisti-cally, previous individual interactions can potentially impact any member of the same “group” during their interactions (e.g. applying for a financial loan, obtain quality hous-ing, job hiring). Importantly, discriminatory interaction, while being negative for the dis-criminated individual, can be positive for other individuals if it is in the context of job hiring, for instance (in-group favouritism, Falk and Zehnder,2007). Thus, social interac-tion can also influence status and/or the percepinterac-tion of it through “Prestige”.

1.6 Experiencing status discrimination

Social psychology helps to explain how mechanisms related to psychosocial stress and interpersonal behaviours associated with status discrimination can aid in understanding the SES-health gradient (Major, Mendes, and Dovidio,2013). During social interactions,

(26)

6 Chapter 1. Background and research goals people automatically assign others to groups based on their age, gender, ethnicity (Major, Mendes, and Dovidio,2013), and social status (Johnson, Richeson, and Finkel,2011). Re-search on prejudice and stereotyping provide evidence that high status individuals often have negative stereotypes and attitudes towards lower status individuals (Major, Mendes, and Dovidio,2013). The feeling of being discriminated, stigmatized or devalued, explicitly or implicitly, is stressful (Major, Mendes, and Dovidio,2013).

Still, individuals that are stigmatized by members of higher status groups are not pas-sive victims of negative stereotypes, attitudes, and discriminatory behaviours. They try to cope with their circumstances.

1.7 Coping with psychosocial stress

Coping is the process of attempting to manage the demands generated by stressful events that are evaluated as exceeding a person’s resources (Lazarsfeld, Merton, et al., 1954). Stressors related to the events will be experienced as stress only if the individual is unable to cope with the event (Miller and Kaiser, 2001). Many different voluntary and invol-untary responses to stress related to status discrimination are known (for extensive de-scription see: Compas et al., 2001) These stress responses are dynamic, multifaceted, and interdependent (Miller and Kaiser,2001). Furthermore, people differ widely in how they respond to stress, and individuals may use different strategies depending on the stressor, and the context in which the event takes place (Miller and Kaiser, 2001). In this regard, Miller and Kaiser,2001state three problems that have appeared in research on stress and coping. The first problem was the increasing number of dimensions characterizing stress and coping responses which are built on poor theoretical and empirical evidence (Miller and Kaiser,2001). The second problem was the mostly unsuccessful identification of in-dividuals coping styles and related adaptive results (Miller and Kaiser, 2001). The third problem is that coping researchers studied stressors ranging from daily struggles, to ma-jor life events and not differentiate between them, which may justify inconsistent findings (Miller and Kaiser,2001). Thus, a large repertoire of possible coping strategies is identified (Compas et al., 2001). Depending on the context, an individual’s psychosocial resources (PSRs) and history of (chronic) stress exposure, the availability and adaptability of coping strategies may vary (Miller and Kaiser,2001; Epel et al.,2018). In this regard, rather than describing different coping strategies, we solely focus on two moments coping strategies come into play and the PSRs that are found to be relevant to influence coping with psy-chosocial stress. The two identified moments are the interpretation of the stressor by the agent – i.e. cognitive appraisal – and recovery from accumulated experienced stress – i.e. recovery.

Cognitive appraisal

Research on stress and coping has found that cognitive appraisals are essential for experi-encing a potentially stressful event as stressful (Miller and Kaiser,2001). How stigmatized people appraise stressors related to discrimination, and which resources are available for coping is critical in understanding how individuals are affected by their stigmatized sta-tus (Miller and Kaiser,2001). Cognitive appraisal happens at the moment an individual perceives a stressor. If the individual appraises the stimulus as non-threatening, because he/she has enough coping resources to manage it, it is not experienced as stressful. How-ever, if the individual does not have enough coping resources, the stimulus is appraised as stressful.

(27)

1.7. Coping with psychosocial stress 7

Recovery

In between stressors, different mechanisms exist that could help individuals to recover from accumulated psychological stress. These include a reappraisal of the stressful event (Jamieson et al.,2018) and recovery over time (Geurts and Sonnentag,2006).

1.7.1 Psychosocial resources

Several PSRs are known to influence coping with psychosocial stressors, and have been consistently linked to favorable mental and physical health outcomes (Taylor and Stan-ton,2007). These include personal mastery, social support and self-esteem (Folkman and Moskowitz, 2004; Scheier and Carver,2003; Taylor and Stanton, 2007). Furthermore, re-search shows that these PSRs are unequally distributed over SES.

Personal mastery

Personal mastery refers to the feeling of being able to control or influence one’s out-comes (Taylor and Stanton, 2007). Individuals of low SES with strong beliefs of mastery have equivalent mental and physical health outcomes compared with persons of high SES groups (Taylor,2011).

Social support

When one has social support, this means a person has people in his or her life that care for that person and will aid in stressful times, if needed (Taylor,2011). Social support can be real (e.g. one receives actual support) or perceived (e.g. one thinks there is support when needed) (Taylor,2011). Research has repeatedly shown that direct social support, and/or the perception of social support, in times of stress, reduces the negative effect of stress, and stimulates psychological adjustment to a large variety of chronically stressful conditions (Taylor,2011).

Self-esteem

Self-esteem, or a positive sense of self, has also proven to be protective against negative mental and physical health outcomes (Taylor, 2011). For example, research has linked a positive sense of self to lower autonomic and cortisol responses to stress (Creswell et al., 2005).

Resource distribution

Individuals with lower levels of SES report less mastery and experience less social sup-port (Adler and Snibbe,2003). The explanation for the unequal distribution of PSRs over SES may be found in the previously mentioned social environmental factors (Evans and Kantrowitz,2002). Increased psychological distress, as a result of exposure to social envi-ronment risk factors, may impair the child’s development of these psychosocial resources, as children are less able to spend time on developing themselves because they need time and cognitive resources to cope with their circumstances (Evans and Kantrowitz, 2002). Furthermore, certain social environments may also hinder the family and their social net-work’s capability from investing time in the development of PSRs in their children (Mack-enbach, 2010). Moreover, their social environments may be less prone to situations in which they experience a sense of mastery, which is needed to construct such a resource.

(28)

8 Chapter 1. Background and research goals All in all, low SES influences individuals’ stress in early life, inducing circumstances that may amplify chronic stress during their lifetime.

1.7.2 Chronic stress

Chronic stress also determines stress responses. Individuals that are experiencing chronic stress have a higher likelihood of perceiving events as stressful and developing maladap-tive stress responses. For example, through the habitual processes (Epel et al., 2018), if individuals have a history of excessive risk exposure, they may anticipate new stressors, which leads to more negative affect resulting in exaggerated appraisals of threat (Epel et al., 2018). Individuals’ cumulative stress exposure thus influences their vulnerability to new stressors.

1.8 Overview

To summarize, in research, the social status of an individual is often determined by mea-suring traditional markers of SES. However, in real life, status cues play an important role in communicating social status. Status is communicated frequently, rapidly and accurately based on non-verbal behaviour, but also through cultural tendencies. During interactions, people judge each other based on their social status. This may be experienced as stressful if individuals are not able to sufficiently cope with stressors caused by this interaction. The ability to cope with stressful situations depends on the availability of PSRs and history of chronic stress, which are known to be unequally distributed over SES. Furthermore, the status discrimination of lower SES individuals may contribute to the existence of status inequality by a general acceptance of their social positions by higher status individuals.

The variety of mechanisms that are related with individual status discrimination dur-ing interaction and the unequal experience of psychosocial stress that results from these interactions indicate that this explanatory pathway likely occurs within a complex social system of interrelated processes that are not sufficiently understood. Analyzing such a complex system is not straightforward. First, there is currently no empirical quantitative data that allows an extensive evaluation of these mechanisms. Second, standard statistical approaches are unsuited for integrating a multitude of feedback and adaptive mechanisms between individual and group level entities over time. Simulation models are more and more recognized as effective tools that can reduce the limitations of standard statistical approaches (Auchincloss et al.,2011).

1.9 Agent-based model approach

Traditional statistical models describing the relation between SES and stress that incorpo-rate a set of statistical assumptions do not take into account the complexity of human be-haviour at the individual level. Therefore, the importance of a mechanism explaining this relation, such as status discrimination, which is the result of interactions between individ-uals with heterogeneous characteristics and behaviour (i.e. SES, PSRs, history of chronic stress, stress responses), might be underestimated. In traditional statistical models, causal-ity is often assumed. These assumptions, however, often ignore feedback loops that may exist between segregated pathways, i.e. vulnerability caused by a history of stressors or the influence of status discrimination on the perception of status members of the same sta-tus group. Decades of public health science may not have found the pathways explaining the robustness of the SES - health gradient because the complexity of human behaviour has often been neglected. To study a phenomenon such as status discrimination in an

(29)

1.10. Research goals 9 isolated fashion is difficult as status discrimination is the result of mechanisms that occur at the individual (micro) level of society and mechanisms that occur at the society level (macro). Furthermore, these mechanisms are influenced by complex constructs of social science and psychology, such as status, discrimination, stress, and coping. To isolate and research the micro and macro mechanisms influencing SES discrimination, we, therefore, propose to use an agent-based model (ABM). ABMs are computer simulations of social interaction between heterogeneous agents, embedded in social structures (Bianchi and Squazzoni, 2015). These models are constructed for observing and analyzing the emer-gence of accumulated outcomes (Bianchi and Squazzoni,2015). Through manipulation of behavioural or interaction model parameters, micro-generative mechanisms that account for macro-scale system behaviour can be explored. The comparative element between agents (during social interactions) with heterogeneous characteristics (SES, PSRs), seems promising in modelling a phenomenon such as SES discrimination.

In the current study, we used an ABM to explore how the unequal distribution of resources (SES, PSRs) contribute to the unequal experience of stress. The model implements inter-action of individuals, the status comparison that occurs during these interinter-actions, and the interpersonal behaviours that come into play when individuals experience stress as a re-sult of status discrimination. The model we implemented was a simple model, not to be used for prediction or to provide policy measures, but rather to be used as an explanatory model that highlights why status discrimination contributes to the SES - health gradient, and the importance of interactions between processes that are related to the phenomenon of status discrimination.

1.10 Research goals

The first goal of this research was to build conceptual models that implement all pre-viously mentioned mechanisms relevant to the status discrimination and psychosocial stress. The second goal was to transform these conceptual models into numerical mod-els which can be used in the ABM. The final goal is to use the ABM to answer the research question: How could individual level status discrimination contribute to growing group level differences in chronic stress?

(30)

(31)

11

Chapter 2

Model building

2.1 Model objective and overview

The main objective of this model is to study how interactions at the individual level could contribute to growing group level differences in chronic stress. The ABM was constructed to be a simple, abstract model that was not intended to be quantitatively calibrated to empirical data, as to our knowledge the data needed for calibration were not available. The model incorporates the social mechanisms related to status discrimination during in-dividual level interactions and the interpersonal mechanisms that are related to dealing with stress resulting from these discriminatory interactions. This qualitative model can be used as a tool for exploring the phenomenon of status discrimination and related stress. The model may 1) act as a basis that could be expanded with more complex behaviours, 2) generate further questions about processes involved, and 3) identify the needs for certain data to be collected for future studies.

In the following sections, we will describe how we constructed a conceptual ABM model and how we operationalised the model numerically. First, we present the data that was used for constructing the agents and the agent population. Second, we describe the type of agents and their characteristics, how we constructed the agent environment, and we present the identified behaviours of the agents and how we numerically imple-mented them into the model. Third, we explain how we impleimple-mented the time steps in the model and how we scheduled interactions and how the model is initialised. There-after, we present the algorithm of the model, which ABM design concepts were used and which outcome measures are taken into account. We shortly describe which tools were used for simulating the model. Then, we show characteristics of the model, including it’s stochastic integrity. Lastly, we describe how we validated the model.

2.2 Data

Two data sources were used while building the model. In section 2.2.1, we present the data used for initialising the agents in the model. In section2.2.2, we present the data that is used for initialling the agent population.

2.2.1 HELIUS

The data for the agents’ characteristics were obtained from the HELIUS (HEalthy Life in an Urban Setting) study (Stronks et al., 2013), a large-scale prospective cohort study being carried out in Amsterdam, the Netherlands. The sample contains Amsterdam res-idents of Surinamese, Turkish, Moroccan, Ghanaian, and ethnic Dutch origin. The study protocol has been described in detail by the authors of the HELIUS study (Snijder et al., 2017); Stronks et al., 2013. In short, baseline data were collected from January 2011 to December 2015. The participants (age 18-70 years) were randomly sampled, through the

(32)

12 Chapter 2. Model building municipality registry of Amsterdam. Note, the sample distribution does not represent the ‘real-world’ distribution of ethnicities in the population of Amsterdam, as minorities were oversampled to establish an even distribution within the HELIUS study. A total of 90,019 individuals were invited to take part in the HELIUS study. Of the invited individuals, 49,952 (55%) responded. Of the respondents, 24,789 (50%) individuals agreed to partici-pate. Noteworthy, the non-response analysis showed that the socioeconomic differences between participants and individuals that chose not to participate were very small (Snijder et al.,2017). The questionnaire was completed by 23,942 (96.6%) participants. The HELIUS study has been approved by the Institutional Review Board of the Academic Medical Cen-ter, University of Amsterdam. Factors being studied include culture, migration history, ethnic identity, socioeconomic status and discrimination. These might affect disease risks through specific pathways, including health-related behaviour and living and working conditions. The variables used from the HELIUS study are presented inTable 2.1.

For the current study, from the total sample of participants who completed the ques-tionnaire (n=23,942), we obtained the data for nearly all (n=22,165). We excluded the par-ticipants who had no full data available on all input- and outcome measures relevant to the ABM. Thus, all participants with missing values for the SES indicators (educational attain-ment, employment status, occupational level and household financial difficulties) and for the PSRs: mastery and social support were excluded. We also excluded participants with an unknown ethnic origin, because we cannot say much about distributions of unknown origins. Thus, leaving only the participants with one of the five ethnicities stated above. In total, 3,934 participants were excluded. Consequently, the analytical sample consisted of 18,231 individuals. The variables of interest are presented inTable 2.1. In the following paragraph, we describe how these variables were obtained in the HELIUS study. How these variables were used for constructing the agent characteristics is described in section

2.3.

The SES variables were reported by the participants of the HELIUS study. The answers from which participants were able to choose from for each of the SES factors are presented in column "values" inTable 2.1.

Mastery was measured using an adapted version of the Pearlin-Schooler Mastery Scale (PMS; Pearlin and Schooler, 1978which has been shown to be reliable in other studies (Slotman et al.,2017). This adapted version uses only the five negatively phrased items of the original PMS (e.g., "I can’t seem to solve some of my problems at all."). The answers were categorised on a 5-point ordinal scale, ranging from 1 (totally disagree) to 5 (totally agree). Items are reverse coded so that higher scores indicate higher degrees of mastery.

The social support measure consists of two parts: the first being Daily Emotional Sup-port (DES; Suurmeijer et al., 1995), a subscale of the Social Support Questionnaire for Transactions (SSQT; Sarason et al.,1983) and the Social Support Questionnaire for Satisfac-tion with the supportive transacSatisfac-tions (SSQS; Helgeson,1993). The DES subscale consists of five items addressing emotional support (e.g., "Do people ever show they understand you?") with a complementary question about the satisfaction of this support (e.g., "Does this happen as much as you’d like?"). The answers were categorised on a 4-point ordinal scale, ranging from 1 ("Hardly ever or never") to 4 ("Often"). As all items are positively phrased, the scores can be used in this study directly.

2.2.2 Centraal Bureau voor de Statistiek

From the Centraal Bureau voor de statistiek (CBS), an autonomous administrative author-ity in the Netherlands, we obtained information about the ethnic distribution in the mu-nicipality of Amsterdam. The data is used to determine the relative prevalence (weight), for each ethnicity found in the HELIUS study, in Amsterdam. These weights were used

(33)

2.2. Data 13

Variable Questionnaire Values

Socioeconomic factors

1: Educational level Completed education (Netherlands, and country of origin) (HELIUS)

2: Never been to school or elementary schooling only 3: Lower vocational schooling or lower sec-ondary schooling 4: Intermediate vocational schooling or intermedi-ate/higher secondary 5: Higher vocational schooling or university Working status Working status (HELIUS) 1: Working

2: Not in the labour force (retired, homemaker, stu-dent)

3: Unemployed / on wel-fare

4: Invalidity Occupational level Occupational level

(HE-LIUS) 1: Elementary professions 2: Lower professions 3: Middle professions 4: Higher professions 5: Scientific professions Financial difficulties Did you experience any

difficulties getting by with your household income? Psychsocial resources

Social support Daily Emotional Support (DES) subscale of Social Support Questionnaire for Transactions (SSQT)

Score: 5 - 20

Perceived social support Social Support Question-naire for Satisfaction with the supportive transac-tions (SSQS-DES)

Score 5 - 20

Mastery Perlin-Schooler Mastery Score: 5 - 25 TABLE2.1: Variables of HELIUS dataset.

(34)

14 Chapter 2. Model building

Ethnicity CBS Ethnicity weights (Amsterdam) HELIUS Population (max N) Total (Ethnicities of HELIUS only) 590777 22117 6863 (= 4564/0.67)

Dutch 392850 0.67 4564 4564 Turkish 44080 0.075 3614 512 Moroccan 76751 0.13 3906 892 Ghanaian 12467 0.021 2339 145 Suriname 64629 0.11 7694 751 Other 470115 47 -Unknown - 1

-TABLE2.2: Distribution of ethnicities in Amsterdam.

for initialisation of the agent population such that the ethnic distribution in the agent pop-ulation resembles the ethnic distribution found in the ’real-world’. Where the data is used for is described in section2.7. The data obtained from CBS, and the determined weights are presented inTable ??.

2.3 Agents

Only a single type of agent was included: an individual inhabitant of the city of Amster-dam. Each agent had a SES and PSRs, both obtained from the HELIUS dataset. Agents with equal SES were members of the same social status group. In sections2.3.1and2.3.2, we describe how we used the information of the SES factors and PSRs obtained from the HELIUS dataset to construct the characteristics of the agents.

2.3.1 Agent SES

For agents to be able to compare each others’ status, they had to be ranked hierarchically based on their SES factors. However, the data used from the HELIUS study for construct-ing this hierarchy is ordinal data - i.e. categorical data where the distances between each category are unknown. Therefore, we can only obtain which agent has "more status" than other agents within each of the SES factors. Moreover, we want the agents to compare on all three SES factors. Introducing a new problem, are education, occupation and wealth equally important for determining status? Or is one factor more important than another? For example, in some parts of the world attained education may be more important then achieved wealth, whilst in other parts it might be reversed. Because little is known on this subjective determination of status, we treated them as equally important for determining SES. By transforming the data of each SES factor into the same scale and sum them with equal weights, we obtained a single score for SES.

The SES composite was constructed using values of the SES indicators: education, occupation, and income. The values for education (edu) and income (inc) were directly obtained from the answers in the HELIUS questionnaire on educational attainment and income difficulty (Table 2.1), as only a single value for these variables is provided. How-ever, for occupation, two values were provided: working status and occupational level. Before we constructed a composite score for SES, we therefore first constructed a compos-ite score for occupation. For this, we used a method described previously in a study using the HELIUS dataset (Snijder et al.,2017).

To construct a composite value for occupation (occ), we first determined the partic-ipant’s working status from the four options (Table 2.1). If the participants were in the

(35)

2.3. Agents 15

Profession Occupation composite score

Scientific 8 Higher 7 Secondary 6 Lower 5 Basic 4 Not in workforce 3 Incapacitated 2 Jobless/ welfare 1

TABLE2.3: Occupation composite scores

Description SES composite sub-score

Occupation score Scientific 6 Higher 5 Secondary 4 Lower 3 Basic 2 Not in workforce 1 Incapacitated 1 Jobless/ welfare 0 Educational attainment

higher vocational schooling or university 6

Intermediate vocational schooling or intermediate/higher secondary schooling (general) 4 Lower vocational schooling or lower secondary schooling 2

Never been to school or elementary schooling only 0

Income difficulty

No, no problems at all 1

No, no problems, but i have to watch what i spend 0.5

Yes, some problems -0.5

Yes, lots of problems -1

TABLE 2.4: SES composite scores for: education, occupation, and income difficulty.

labour force, we further determined their occupational level (Table 2.1). By doing so, par-ticipants may have a composite occupation value (from now on occupation score) between 1 and 8. The ranking of the scores is presented inTable 2.3.

Using the three SES indicators, we constructed the SES composite. First, we translated the value from the HELIUS to a score used in calculating the composite. The translations are presented inTable 2.4. Note that the scores for "incapacitated" and "not in workforce" are the same, and the score of "jobless/welfare" is set to 0, such that the maximum occu-pation score is 6. Furthermore, the scores for the education options are multiplied with 2, such that the maximum scores for occupation and education are the same. Because income difficulty is a subjective measurement, it’s score is used to add a penalty or award a bonus to the SES score ranging from -1 to 1. Equation 2.1is used to calculate SES of agents (As).

For clarity, the weights (α1−3) are thus treated as equal. However, these may be adapted

if more knowledge is obtained on the importance of each SES factor in the determination of social status. An additional two points were added to the final score such that the min-imum status is set to 1. In doing so, we obtained a granularity of 29 positions in the SES

(36)

16 Chapter 2. Model building hierarchy ranging from 1 to a maximum of 15.

As=2+ (α1·edu+α2·inc+α3·occ) (2.1)

2.3.2 Agent PSRs

As described in section1.7.1, several PSRs are known to influence coping with psychoso-cial stress. Because the HELIUS study provided values for Mastery and Sopsychoso-cial support, obtained using scientifically validated questionnaires, we used these two. Studies on the buffering effects of PSRs on stress differentiate in the effectiveness of each resource on stress in various contexts (Miller and Kaiser, 2001; Epel et al.,2018; Taylor and Stanton, 2007). However, many studies agree that PSRs in general are related to efficient coping (Miller and Kaiser, 2001; Epel et al., 2018; Lazarsfeld, Merton, et al., 1954). Therefore, in this model, both values were combined to obtain one score for PSRs.

To obtain a single score for PSRs for each of the participants, we first normalised each of the resources by scaling the values between 0 and 1 for all participants. Thereafter, the two values are summed. And finally, the summed score is again normalised by scaling between 0 and 1.

2.4 Environment

The environment of the agents is a social network. Since information on social ties be-tween participants of the HELIUS study was not reported, the social network structure had to be estimated. To estimate the social ties between agents, we, therefore, used a ba-sic organising principle of social network science: the homophily principle (McPherson, Smith-Lovin, and Cook,2001).

The homophily principle describes the tendency of individuals to have network ties with similar other (McPherson, Smith-Lovin, and Cook, 2001). These ties may include marriage, friendship, work, advice, support, information transfer, and the more superfi-cial ties in an individual’s network (McPherson, Smith-Lovin, and Cook, 2001). In this regard, research shows that individuals’ social networks are largely homogeneous with regard to socioeconomic and sociodemographic characteristics (McPherson, Smith-Lovin, and Cook, 2001). Therefore, the information people receive, the attitudes they form, and the interactions they experience are bounded by homophily.

Previous research has distinguished two types of homophily: value- and status ho-mophily (Lazarsfeld, Merton, et al., 1954). In value homophily similarity is based on values, attitudes, and beliefs and includes a variety of internal states that are assumed to define people’s orientation of future behaviour (McPherson, Smith-Lovin, and Cook, 2001). In status homophily, similarity is based on the sociodemographic and socioeco-nomic dimensions, including ascribed characteristics such as ethnicity, gender or age, and acquired characteristics such as education, occupation, wealth, and behavioural patterns (McPherson, Smith-Lovin, and Cook,2001).

Given that we only have data available on sociodemographic and socioeconomic di-mensions, and this study researches status discrimination, we operationalised similarity between agents based on the principle of status homophily (as opposed to value ho-mophily). In this regard, we constructed a social network where relational strength be-tween agents is determined by their similarity on the socioeconomic factors: educational attainment and occupational score. We excluded the socioeconomic factor financial diffi-culties from the similarity calculation as individuals may experience financial diffidiffi-culties

(37)

2.5. Behaviors 17 regardless of how much money they earn. We assume this does not influence the proba-bility of interaction.

The social network of the agents is operationalised as a complete graph, where the nodes represent the agents, and the edges represent the relational strength between agents. This relational strength is determined by the agents’ similarity (e.g. homophily). The probability of two agents interacting will depend on their relational strength.

Similarity scores between agents (Di,j) were calculated using Equation 2.2. The

Eu-clidean distance (d) between agents, is calculated using educational attainment and oc-cupational score as Cartesian coordinates. We included a parameter which controls the baseline similarity (σ), which thus determines the minimal similarity between agents.

p= (edui, occi) q= (eduj, occj) d(p, q) = v u u t 2

∑

i=1 (qi−pi)2 Di,j =Dj,i=1/d(p, q) +σ (2.2)

2.5 Behaviors

Using the literature presented in chapter1and section2.4, we identified a set of behaviours that are associated with the process of status discrimination. Two types of behaviours were identified: behaviours associated with the social processes, and behaviours related to the psychological process. Social behaviour includes, how agents interact with other agents, how agents estimate status of other agents, how they discriminate on the basis of this status estimation, and how agents gain/lose prestige because of discrimination. Psychological behaviour includes, how agents perceive stressors resulting from status dis-crimination, and how agents cope with the perceived stressor, how the accumulation of experienced stress causes individuals to become vulnerable to new stressors. In the fol-lowing sections, each of these behaviours is conceptualised and translated into numerical functions which were used in the ABM.

2.5.1 Interact

The chance of having a certain interaction partner is based on the information presented in 2.4. During an interaction event, the agent has to interact with another agent, where the probability of selection depends on the similarity with other agents. The higher the similarity, the higher the probability of interacting. Note, within the same time step agents only meet a single time, e.g. cannot choose each other as interaction partner in the same time step. The choice of an agent to interact with another is operationalised as a simple probability-based selection.

2.5.2 Estimate status

During an interaction, agents estimate each others’ status. The literature used to construct this behaviour is described in section1.5. The estimation is based on the agent SES (As)

and the agent’s status group prestige (Gpr). Individuals communicate their status through

status cues. However, individuals may adapt to different contexts by using different status cues. Furthermore, the estimation of status is not exact. Therefore, estimating an agents

(38)

18 Chapter 2. Model building status is subjected to an estimation error (εs,t)both for the perceiver and the

communica-tor. The numerical implementation of status estimation is presented inEquation 2.3. The estimation error is pulled randomly from a uniform distribution which is bounded by the parameter status estimation noise (εs). We chose a uniform distribution as we assume this

error is independent of the actual status. If an estimated status is below 0, it is set to 0. Ei = Ais+Gpri +U(−εs, εs)

Ej = Asj+Gprj +U(−εs, εs)

IF E<0, E=0.

(2.3)

2.5.3 Discriminate

Based on the information presented in1.6, we operationalised a process of discrimination. The agent with the lowest estimated status is discriminated. In the following equations, we assumed Ai to be discriminated and Ajto be the discriminator.

2.5.4 Perceiving the stressor

The perceived stressor is determined by translating the amount of status difference into a stressor (s). Furthermore, based on the information presented in1.7.2, the perception of stress is influenced by previous experiences of stress which may cause individuals to per-ceive stressors as more severe. The discriminated agent perper-ceives the stressor. The severity of the stressor depends on the perceived status difference (St_d) and the agent’s vulnerabil-ity (Av). The amount of stress is controlled using the stressor coefficient parameter (ξ). We

assumed that small status differences cause less stress and that there exists a maximum amount of stress caused by status difference.

In this regard, the numerical function f(x)(Equation 2.4,Figure 2.1) that translates the perceived status difference to an experienced stressor is operationalized as a sigmoidial function. This sigmoidal function is bounded such that St_d = 0, f(St_d) = 0 and St_d >

20, f(Std) =1 with a weak inflection point at Std=7.5. Similar to to this, we assumed that

vulnerability increases slowly for small amounts of stress of accumulated stress, and that there exists a maximum vulnerability if an agent has maximum accumulated stress. Fur-thermore, the amount of vulnerability is controlled with the vulnerability coefficient (υ). The agent’s Viis obtained by transforming Siusing function g(x)(Equation 2.5,Figure 2.2).

This sigmoidal function is bounded such that S =0, f(S) = 0 and S>=1, f(S) = 1 with a inflection point at S=0.4. f(x) = −0.06739+ 1.08 (1+e−0.35(x−7.5)₎ Std = |Ei, Ej| s= f(Std) · (1+Aiv) ·ξ (2.4) g(x) = −0.01798+ 1.0 (1+e−10(S−0.4)₎ Av= g(S) ·υ (2.5) 2.5.5 Cope

The literature relevant to the coping mechanisms is presented in1.7. Being discriminated against will be perceived as stressful only if the event poses a threat that could exceed

(39)

2.5. Behaviors 19

FIGURE2.1: Function for translating status difference to perceived stressor

FIGURE2.2: Function for translating accumulated stress to vulnerability

an individual’s PSRs (Miller and Kaiser, 2001). The coping process is conceptualised as the first moment where coping strategies come into play, namely cognitive appraisal (de-scribed in section 1.7). The efficiency of this process depends on the available PSRs. We operationalised this process numerically using a linear relation, where the amount of ac-tually experienced stress (s0) depends on the severity of the stressor (s), the amount of available coping resources (Ar), and the coping noise (εr). We introduce coping noise, for

similar reasons as the status estimation error described in section2.5.2. Depending on the situation, the efficiency of coping responses might differ. However, this error is pulled from a normal distribution, as we assumed that the efficiency of coping is dependent on the agent’s actual coping resources. The amount of available coping resources is controlled with the PSRs coefficient (ρ). The numerical implementation is presented inEquation 2.6.

s0 =s−Arρ+ N (0, εr)

IF s0 <0, s0 =0 (2.6)

2.5.6 Experience stress

The agent experiences stress for the amount of stressor that is exceeding the amount of coping resources. Stress accumulates (S) if, in previous timesteps, stress is experienced. The equation for this process is described in Equation 2.6. We have seen in section 2.5.4

(40)

20 Chapter 2. Model building that this accumulated stress influences the vulnerability to new stressors.

Si,t=Si,t−1+s0 (2.7)

2.5.7 Gain/lose prestige

Based on the information presented in section1.5.4, we conceptualised the prestige mech-anisms in the following way. Because of interaction, either the discriminated agent’s status group loses prestige, or the discriminating agent’s status group gains prestige. Whether it is the discriminated or discriminating agent is chosen randomly. The rate of change depends on the prestige growth rate (Prβ). However, because of the unequal amount of agents in each status group, some status groups have more interactions in which prestige may be updated than other status groups. Therefore, the prestige growth rate is corrected for the number of agents belonging to that group (Gn). Furthermore, we assumed that

gaining prestige becomes harder when belonging to the top prestigious ranks of a pop-ulation and that there exists a maximum prestige. Therefore, the prestige growth rate is bounded at Gpr =30, and depends on the already achieved prestige. The numerical

imple-mentation of the prestige mechanisms is presented inEquation 2.8, where h(x)represents the function that bounds Ptβsuch that the maximum achievable prestige is 30. Note, that distance (1/Gpr) inEquation 2.8increases (+) if the agent is discriminated and decreases

(-) if an agent discriminates. h(x) =1.5− 1.5 (1+e−0.225(Gpr−3)₎ Prβ =h(Gpr) G0_pr=1/( 1 Gpr · (1± Prβ Gn )) (2.8)

FIGURE2.3: Function for determining the prestige growth rate coefficient

2.5.8 Recover

As described in section1.7, the second moment coping strategies come into play was be-tween experiences of stress. We assumed that recovery of accumulated stress, for example, in between stressors, is independent of any resource. We operationalised this by reducing the amount of accumulated stress (S) with a fixed recovery rate (Rβ) at the end of each

Modelling the influence of individual level status discrimination on group level differences in chronic stress. Agent based model approach

U

NIVERSITEIT VAN

A

MSTERDAM

& V

RIJE

U

NIVERSITEIT

A

MSTERDAM

M

T

Modelling the influence of individual level

status discrimination on group level

differences in chronic stress

An Agent-Based model approach

Declaration of Authorship

Abstract

Acknowledgements

Contents

List of Figures

List of Tables

List of Abbreviations

List of Symbols

Chapter 1

Background and research goals

1.1

The socioeconomic status – health gradient

1.2

Mechanisms related to SES - health gradient

1.3

Mechanisms related to stress

1.4

Status discrimination and psychosocial stress

1.5

Social status evaluation

1.6

Experiencing status discrimination

1.7

Coping with psychosocial stress

1.8

Overview

1.9

Agent-based model approach

1.10

Research goals

Chapter 2

Model building

2.1

Model objective and overview

2.2

Data

2.3

Agents

2.4

Environment

∑

2.5

Behaviors